Downscaling the variable tas

• Climate response variable: - sfcWind, tas, pr: original scale daily, reduced scale: hourly - clt, rsds: original scale monthly, reduced scale: daily

• Training: - Response: ERA5 in reduced scale - Data perdiod: 1980 - 2015.
  - Predictors: from CMIP6 (daily or monthly) - Predictors variables: sfcWind, tas, pr, tasmax, tasmin and psl in daily scale, clt, rsdt,rsds in monthly scale, the month , hour, sun’s elevation & azimuth and the daily daylight amount in seconds.

• Validation:

  • Period: 2015-01-02 00:00:00 to 2023-12-31 23:00:00
  • Variable to predict: reduced scale CMIP6 variable

Results

cesm2.ssp245

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	−0.43	−0.51	−0.53	−0.62
ratio_of_sd	0.96	1.00	0.95	0.93
amplitude_ratio_of_means	0.75	0.96	0.82	0.72
maximum_error	0.44	0.13	0.26	0.27
ks_mean_on_coarse_res_with_extremes	0.29	0.14	0.17	0.28
qqplot_mae	1.26	1.48	1.54	1.81
acf_mae	0.13	0.03	0.08	0.11
extremogram_mae	0.08	0.04	0.02	0.04

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

cesm2.ssp370

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	−0.50	−0.57	−0.59	−0.68
ratio_of_sd	0.93	0.97	0.92	0.89
amplitude_ratio_of_means	0.74	0.96	0.81	0.71
maximum_error	0.42	0.13	0.25	0.27
ks_mean_on_coarse_res_with_extremes	0.21	0.04	0.12	0.22
qqplot_mae	1.45	1.65	1.74	1.97
acf_mae	0.12	0.02	0.07	0.10
extremogram_mae	0.04	0.02	0.02	0.02

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

cesm2.ssp585

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	−0.37	−0.44	−0.47	−0.56
ratio_of_sd	0.96	0.99	0.95	0.92
amplitude_ratio_of_means	0.74	0.96	0.81	0.71
maximum_error	0.41	0.11	0.25	0.26
ks_mean_on_coarse_res_with_extremes	0.17	0.05	0.09	0.19
qqplot_mae	1.08	1.29	1.38	1.63
acf_mae	0.13	0.03	0.08	0.11
extremogram_mae	0.06	0.02	0.01	0.03

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

ec_earth3.ssp434

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	−0.19	−0.27	−0.28	−0.39
ratio_of_sd	0.97	1.02	0.96	0.93
amplitude_ratio_of_means	0.67	0.95	0.72	0.62
maximum_error	0.45	0.16	0.26	0.34
ks_mean_on_coarse_res_with_extremes	0.37	0.18	0.28	0.37
qqplot_mae	0.56	0.77	0.84	1.14
acf_mae	0.17	0.07	0.14	0.17
extremogram_mae	0.07	0.02	0.03	0.05

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

mri_esm2_0.ssp245

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	−0.15	−0.32	−0.34	−0.41
ratio_of_sd	1.14	1.15	1.09	1.09
amplitude_ratio_of_means	0.83	0.96	0.92	0.82
maximum_error	0.44	0.11	0.22	0.30
ks_mean_on_coarse_res_with_extremes	0.42	0.30	0.29	0.35
qqplot_mae	0.77	1.03	0.98	1.20
acf_mae	0.14	0.09	0.10	0.12
extremogram_mae	0.11	0.10	0.02	0.02

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

mri_esm2_0.ssp370

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	0.03	−0.13	−0.17	−0.24
ratio_of_sd	1.09	1.11	1.06	1.04
amplitude_ratio_of_means	0.79	0.96	0.88	0.77
maximum_error	0.43	0.11	0.22	0.32
ks_mean_on_coarse_res_with_extremes	0.43	0.29	0.30	0.40
qqplot_mae	0.55	0.63	0.54	0.70
acf_mae	0.14	0.07	0.10	0.12
extremogram_mae	0.07	0.03	0.02	0.04

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.
  

mri_esm2_0.ssp434

Numerical evaluation metrics

• Each row correspond an evaluation metric
• Dark blue cell indicates the best model in the row.
• Light blue cell indicates the second best model in the row.

Metric	cnn	naive	xgboost	lstm
diff_of_means	0.00	−0.17	−0.20	−0.27
ratio_of_sd	1.14	1.15	1.10	1.09
amplitude_ratio_of_means	0.82	0.96	0.90	0.80
maximum_error	0.44	0.11	0.22	0.32
ks_mean_on_coarse_res_with_extremes	0.41	0.28	0.30	0.37
qqplot_mae	0.76	0.88	0.72	0.78
acf_mae	0.15	0.09	0.11	0.13
extremogram_mae	0.08	0.05	0.02	0.02

Time series of the first few days

Graphical evaluation metrics

• Observations: cumulative distribution (Detrended QQ), dependency structure (ACF), and extremes (Extremogram, using 0.97 cut off).

• Daily/Monthly gruped: Maximum hour histogram (only in daily variables), Amplitude distribution.

• Hourly/Daily grouped: Mean profile per hour within days (or weeks witihin season)

• In all plots, grey color represent the true ERA5 signal.